Abstract
Parking during large-scale events often leads to systemic traffic failures and significant "cruising" time. This paper presents a smart navigation system that accounts for heterogeneous driver characteristics—varying strategies and patience levels—to optimize parking assignments. By employing a Mixed Search Algorithm, the system reduces average parking search time to 4.7 minutes, outperforming traditional single-strategy benchmarks.
Introduction
Urban parking congestion is a persistent issue, particularly during major events like concerts or sports games. Research indicates that up to 30% of urban traffic consists of drivers searching for parking, leading to increased fuel consumption and emissions. Conventional parking guidance systems often assume uniform driver behavior, typically directing everyone to the lot closest to the destination. However, real-world drivers exhibit diverse priorities: some prioritize proximity to the event, while others seek to avoid congestion or minimize driving distance from their entry point. This study addresses these complexities by modeling diverse human driving behaviors within a unified optimization framework.
Methods
The proposed system utilizes a parking assignment optimization method designed to minimize the total time spent by drivers. This total time includes driving to a lot ($TD_{ik}$), searching for an available spot within that lot ($TS_{ik}$), and walking to the final destination ($TW_i$).
The core of the approach is the Mixed Search Algorithm, which categorizes drivers into four distinct groups:
- Group 1: Minimizes driving distance from the city entry point.
- Group 2: Minimizes driving distance while avoiding high-demand lots adjacent to the stadium.
- Group 3: Minimizes walking distance to the event.
- Group 4 (Comprehensive): Minimizes the sum of all travel components (driving and walking).
Additionally, the model incorporates a driver tolerance threshold modeled via a gamma distribution. This allows the simulation to account for "failed searches," where drivers abandon the search area after encountering multiple full lots—a realistic behavior often overlooked in standard models.
Datasets and Experimental Setup
The study's simulation is grounded in a real-world geographic dataset centered on California Memorial Stadium in Berkeley, CA.
The infrastructure data includes 21 visitor parking lots with a total capacity of 3,992 spaces. To simulate event-day conditions, an arrival stream of 3,992 vehicles was generated over a two-hour window (10 AM to 12 PM) using a Poisson distribution with Gaussian noise. Spatial coordinates were converted using Miller projections to facilitate precise Manhattan distance calculations for walking and driving paths.
Key Results and Insights
By acknowledging driver heterogeneity, the system naturally distributes traffic more effectively across the available parking inventory. This prevents the bottlenecks typical of "nearest-lot" strategies. The Mixed Search approach was evaluated against two established baselines:
- Shin’s Method (Nearest to destination): Average rerouting time of 14.6 minutes.
- Rehena’s Method (Total travel time optimization): Average rerouting time of 6.1 minutes.
- Proposed Mixed Search: Achieved a superior average rerouting time of 4.7 minutes.
Qualitative analysis revealed that while single-strategy methods often left peripheral lots underutilized and caused "failed search" spikes at central locations, the mixed search approach ensured balanced utilization across all 21 lots, including more distant locations like lots 8 and 12.
Limitations and Future Work
The current simulation does not yet fully capture behavioral complexities such as illegal parking or mid-route decision changes based on visual cues. Future research aims to integrate real-time traffic sensor data and IoT infrastructure to enable dynamic, real-time assignments. Validating these models with empirical data from actual event days remains a critical next step to bridge the gap between simulation and curb-side reality.
Summary updated